Analyzing Quantum Circuits Via Polynomials
Analyzing Quantum Circuits Via Polynomials
Kenneth W. Regan1
University at Buffalo (SUNY)
15 November, 2013
1
Includes joint work with Amlan Chakrabarti and Robert Surówka
Analyzing Quantum Circuits Via Polynomials
Quantum Circuits
Quantum circuits look more constrained than Boolean circuits:
But Boolean circuits look similar if we do Savage’s TM-to-circuit
simulation and call each column for each tape cell a “cue-bit.”
Analyzing Quantum Circuits Via Polynomials
Analyzing Quantum Circuits Via Polynomials
Analyzing Quantum Circuits Via Polynomials
Analyzing Quantum Circuits Via Polynomials
Analyzing Quantum Circuits Via Polynomials
Toffoli Gate
Analyzing Quantum Circuits Via Polynomials
Bounded-error Quantum Poly-Time
A language A belongs to BQP if there are uniform poly-size quantum
circuits Cn with n data qubits, plus some number α ≥ 1 of “ancilla
qubits,” such that for all n and x ∈ { 0, 1 }n ,
x∈A
=⇒
Pr[Cn given hx0α | measures 1 on line n + 1] > 2/3;
x∈
/A
=⇒
Pr[. . . ] < 1/3.
One can pretend α = 0 and/or measure line 1 instead. One can also
represent the output as the “triple product” ha | C | bi, with a = x0α ,
b = 0n+α .
Two major theorems about BQP are:
(a) Cn can be composed entirely of Hadamard and Toffoli gates [Y.
Shi].
(b) Factoring is in BQP [P. Shor].
Analyzing Quantum Circuits Via Polynomials
What is Known About BQP?
BPP ⊆ BQP.
BQP ⊆ PP [Adleman-Demarrais-Huang, 1998]
The acceptance probability px of a QC on input x can be written as
px =
f (x) − g(x)
√
2m
where f and g are #P functions whose nondeterminism ranges over
m = nO(1) binary variables. Hence [Fortnow-Rogers, 1999] BQP is
in a class AWPP ostensibly weaker than PP.
BQP is not known to include graph-isomorphism or MCS
(min.-circuit size).
Analyzing Quantum Circuits Via Polynomials
Translation Into Polynomials
Dawson et al. [2004] showed that for QC’s of Hadamard and Toffoli
gates, f and g could be the functions counting solutions to two sets
E1 and E0 of polynomial equations over Z2 .
Applied by Gerdt and Severyanov [2006] to build a
computer-algebra simulation of these quantum circuits.
[This talk] We make E1 and E0 each a single equation, over any
desired field or ring, with direct translation of a much wider set of
quantum gates, without resporting to mixed-modulus arithmetic.
Some motivations:
Build more extensive simulations—Chakrabarti.
Understand which QC’s can be simulated “classically.”
Idea for putting BQP into the third level of PH, via Stockmeyer’s
approximate counting.
Ideas for algebraic metrics of multi-partite entanglement.
Limitations on scalability of QC’s?
Analyzing Quantum Circuits Via Polynomials
Target Rings
Given a QC C, define k(C) to be the least integer such that all
phase angles of gates in C are multiples of 2π/k.
A ring is adequate for C if it embeds the k-th roots of unity, either
multiplicatively or additively.
Also embed e(0) = 0 in the multiplicative case (“p-case”) and
e(0) = a set of dummy variables w in the additive case (“q-case”)
(a key trick, given below).
For Toffoli+Hadamard, k = 2, and Dawson et al. gave an additive
embedding into Z2 . Whereas the p-case needs Z3 inside the field, so
−1 6= +1.
For the T -gate which has entries eπi/4 , k = 8.
The gates in Shor’s QFT circuits have large k. But, they can be
approximated by circuits with Hadamard and Toffoli only, with
k = 2!
Analyzing Quantum Circuits Via Polynomials
Polynomials and Equation Solving
We will translate quantum circuits with n lines and s gates. Each
interior juncture is denoted by a variable zij (1 ≤ i ≤ n; 1 ≤ j ≤ s − 1).
A gate is balanced if all non-zero entries in its gate matrix have the
same magnitude r.
All the most prominent gates are balanced.
Given a QC of balanced gates, let R be the product of the
balancing magnitudes r over its gates.
For a polynomial p in variables ai , bi , zij and arguments
a, b ∈ { 0, 1 }n , pa,b denotes the polynomial in variables zij resulting
from substituting the arguments.
NB [pa,b (zij ) = v] denotes the number of binary solutions to the
equation, i.e. with an assignment from { 0, 1 }n(s−1) to the zij
variables.
Now we can state the theorem for the multiplicative case.
Analyzing Quantum Circuits Via Polynomials
Main Theorem—Multiplicative Case
Theorem
There is an efficient uniform procedure that transforms any balanced
n-qubit quantum circuit C with s gates into a polynomial p such that for
all a, b ∈ { 0, 1 }n :
ha| C |bi = R
k−1
X
ω ` NB [pa,b (zij ) = e(ω ` )]
(1)
`=0
over any adequate ring. The size of p as a product-of-sums-of-products
of zij and (1 − zij ) is O(22m ms) where m is the maximum arity of a gate
in C, and the time to write p down is the same ignoring factors of log n
and log s for variable labels.
Analyzing Quantum Circuits Via Polynomials
Main Theorem—Additive Case For Zk
Theorem
There is an efficient uniform procedure that transforms any balanced
n-qubit quantum circuit C with s gates, whose nonzero entries have
phase a multiple of 2π/k for k a power 2r , into a polynomial q(~a, ~b, ~z, w)
~
over Zk such that for all a, b ∈ { 0, 1 }n :
ha| C |bi = Rk
−s
k−1
X
ω ` NB [qa,b (zij , wsj ) = e(ω ` )],
`=0
with R and the size of q the same as for p in Theorem 1.
(2)
Analyzing Quantum Circuits Via Polynomials
Substitution and Nondeterminism
When there is no gate between junctures j − 1 and j on qubit line
i, or if the gate in column j leaves qubit i unchanged (as with a
control), then one can substitute:
i
zji = zj−1
.
Thus a new internal variable is introduced only when one cannot
substitute.
This happens with Hadamard gates.
Nondeterminism = the number of internal variables.
P 0 denotes polynomials obtained from the P formally given by
Theorem 1 by substitution.
Q0 likewise from Q in Theorem 2.
P 00 denotes a particular embedding into the ring Z2 [u] where the
adjoined element u satisfies u4 = 1, so it translates i.
Analyzing Quantum Circuits Via Polynomials
Examples of Gate and Circuit Simulations
Projected from a draft of the paper. . .
Definition. Two polynomials are equivalent if they arise from
annotations of two equivalent quantum circuits.
Analyzing Quantum Circuits Via Polynomials
Annotating a circuit
x1
a1
H
a2
H
a3
b2
x3
•
H
•
x1 b2 + a3 − 2x1 b2 a3
b3
H
b2 b3 + x3 − 2b2 b3 x3
b1
•
b2
•
b3
Analyzing Quantum Circuits Via Polynomials
What to do with all this—in theory?
Two central theoretical problems are:
(1) Which subsets of quantum gates can be simulated efficiently with
classical computation alone?
(2) What (classical) upper and lower bounds can be given for BQP?
Both problems involve one in subcases of the classic #P-complete
problem of counting solutions to polynomial equations. Unlike the case
of SAT, there has not been a comparable classification theorem, though
Leslie Valiant and Jin-Yi Cai and their students have undertaken one.
Analyzing Quantum Circuits Via Polynomials
Case of Stabilizer Circuits
Are QC’s with only Hadamard, S, and cnot and/or CZ gates.
Have efficient classical simulations: O(s3 ) by Gottesmann-Knill,
O(s2 ) by Aaronson-Gottesmann, O(s) by Peter Hoyer
(give-and-take log n factors).
Additive translation into equations over Z4 :
1
2
3
4
Hadamard: 2yz, with no substitution; and
S: y 2 , substituting z := y; and
CZ : 2y1 y2 , substituting z1 := y1 , z2 := y2 ; or
cnot: 0, substituting z1 := y1 , z2 := y1 + y2 , with the latter being
sound in place of the proper z2 := y1 + y2 − 2y1 y2 owing to the
invariance under adding 2.
Analyzing Quantum Circuits Via Polynomials
Yet Another Proof of Dequantization
Theorem (Cai-Chen-Lipton-Lu, 2010)
Quadratic n-variable polynomials over Z2r for fixed r have
polynomial-time solution colunting.
Open for variable r = nO(1) .
Corollary
The exact acceptance probability for stabilizer circuits can be computed
in deterministic polynomial time.
General running time from CCLL is inferior to best-known “graph
state” methods for stabilizer circuits. Can this be matched for the
particular polynomials we get over Z4 ?
Analyzing Quantum Circuits Via Polynomials
Graininess of Solution Set Sizes
Theorem (Surówka)
Let P (x) be a multivariate polynomial of n variables over Zm where
m = pr11 pr22 . . . prkk and all p1 , p2 , . . . , pk are different primes. Then for
any g ∈ Zm there is an integer Tg such that:
NP [g] = Tg
Y
i:2|ri
ri
(n−1)
pi2
Y
i:2-ri
ri −1
(n−1)
2
pi
Analyzing Quantum Circuits Via Polynomials
Graininess of Solution Set Sizes
Theorem (Surówka)
Let P (x) be a multivariate polynomial of n variables over Zm where
m = pr11 pr22 . . . prkk and all p1 , p2 , . . . , pk are different primes. Then for
any g ∈ Zm there is an integer Tg such that:
NP [g] = Tg
Y
i:2|ri
ri
(n−1)
pi2
Y
ri −1
(n−1)
2
pi
i:2-ri
Proof applies Hensel lifting. But we believe we can go beyond what
Hensel’s techniques, as used by Ax and others, give.
Analyzing Quantum Circuits Via Polynomials
Beyond Lifting...
Also in terms of the degree, we conjecture the following stronger result,
with supportive computer runs:
Conjecture
Let P (x) be a multivariate polynomial of degree d, of n variables over
Zpr1 pr2 ...prk where all p1 , p2 , . . . , pk are different primes. Then for any
1 2
k
g ∈ Zpr1 pr2 ...prk there is an integer Tg such that:
1
2
k
NP [g] = Tg
Y
i:ri =1
d n e−1
pi d
Y
i:ri >1
d
pi
ri n
e−1
2
.
Analyzing Quantum Circuits Via Polynomials
The Other Goals—Ideas Welcome
Extend notion of equivalence to manipulations giving polynomials
that do not come from QC’s.
Try to increase the R factor without introducing more
nondeterminism. That makes Stockmeyer approximation “better.”
What notions from algebraic geometry might yield measures of
entanglement?
Idea: It should reflect constraints on solution spaces. This aligns it
with the idea of geometric degree of algebraic varieties.
Ultimately goal is to apply Strassen’s lower-bound ideas to QC’s.